Automatic Detection of Contrast Injection

1. A method for automatically detecting the presence of a contrast agent in an x-ray image, comprising: acquiring a preliminary x-ray image of a region of interest of a subject prior to administration of the contrast agent; estimating a background image based on the acquired preliminary x-ray image; administering the contrast agent into the subject; acquiring a main set of x-ray images including a plurality of image frames; subtracting the estimated background image from each image frame of the acquired main set of x-ray images to create a plurality of subtracted images corresponding to the plurality of image frames; determining a measure of image intensity for each of the subtracted images; selecting one or more of the subtracted images having a highest image intensity; fitting a predefined shape model to the selected one or more subtracted images by using a semi-global optimization strategy; using the fitting of the predefined shape model to the one or more subtracted images to fit the shape model to each of the plurality of subtracted images; calculating a feature value for each image frame based on pixel intensities of each pixel fitted to the shape model for the corresponding subtracted image; determining an image frame of peak contrast by selecting the image frame with the greatest feature value; and using the determined image frame of peak contrast to indicate the presence of the contrast agent in the main set of x-ray images, wherein the semi-global optimization strategy comprises: performing a global search on a 2-D space of translation with a course grid using three different groups of five scale and rotation parameters to find a course-grid optimization; starting from the course-grid optimization, performing a global search with all five scale and rotation parameters within a region defined by a size of one grid of the course grid to enhance the course-grid optimization; and performing fine-tuning of the enhanced course-grid optimization using a hill climbing algorithm as a local optimization strategy within a region that is smaller than the size of one grid of the course grid, wherein the five scale and rotation parameters include x-direction scale, y-direction scale, x-direction translation, y-direction translation, and rotation.

2. The method of claim 1 , additionally comprising detecting a probe from within the preliminary x-ray image and generating a probe mask therefrom, and in calculating the feature value for each image frame based on pixel intensities of each pixel fitted to the shape model for the corresponding subtracted image, pixels corresponding to the generated probe mask are excluded.

3. The method of claim 1 , wherein the estimation of the background is updated as the main set of x-ray images are acquired based on one or more most recent image frames that are classified as not including contrast.

4. The method of claim 1 , wherein the measure of image intensity for each subtracted image includes generating a non-linear histogram mapping of each subtracted image.

5. The method of claim 4 , wherein selecting one or more of the subtracted images having a highest image intensity includes determining one or more highest histogram values.

6. The method of claim 1 , wherein the predefined shape model is modified in accordance with an anatomical structure of the subject prior to fitting.

7. The method of claim 1 , wherein the predefined shape model represents an aortic root.

8. The method of claim 1 , additionally comprising: selecting a set of contrast frames from among the plurality of image frames that have a feature value that is sufficiently close to the frame of peak contrast; selecting a set of non-contrast frames from among the plurality of image frames that have a feature value that is sufficiently far to the frame of peak contrast; training a local classifier using the selected set of contrast frames as positive training data and using the selected set of non-contrast frames as negative training data; and determining whether each of the frames of the plurality of image frames that are neither sufficiently close to the frame of peak contrast nor sufficiently far to the frame of peak contrast are contrast frames or non-contrast frames using the trained local classifier.

9. The method of claim 1 , additionally comprising registering a 3D image of an aortic root to the image frame determined to be of peak contrast and displaying the registered image.

10. The method of claim 9 , wherein the displayed registered image is used as visual guidance in performing an interventional procedure.

11. A method for automatically detecting a contrast agent in an x-ray image, comprising: acquiring a preliminary x-ray image of a region of interest of a subject known to exclude the contrast agent; detecting a probe from within the preliminary x-ray image and generating a probe mask therefrom; estimating a background image based on the acquired preliminary x-ray image; acquiring a first set of x-ray images including a plurality of image frames; subtracting the estimated background image from each image frame of the acquired first set of x-ray images to create a plurality of subtracted images corresponding to the plurality of image frames; determining a measure of image intensity for each of the subtracted images; selecting one or more of the subtracted images having a highest image intensity; comparing each of the selected images with the estimated background image and determining that the first set of x-ray images does not include the contrast when each of the selected images are within a predetermined measure of similarity to the background image, and when at least one of the subtracted images exceeds the predetermined measure of similarity to the background image, the following additional steps are performed: fitting a predefined shape model to the selected one or more subtracted images using a semi-global optimization strategy; using the fitting of the predefined shape model to the one or more subtracted images to fit the shape model to each of the plurality of subtracted images; calculating a feature curve for set of x-ray images based on pixel intensities of each pixel fitted to the shape model for the corresponding subtracted image while excluding pixels corresponding to the generated probe mask; performing frequency analysis on the calculated feature curve to identify a case in which high contrast feature value is attributable to cardiac or respiratory motion; and when it is identified that the first set of x-ray images does not have a high contrast feature value attributable to cardiac or respiratory motion, the following additional step is performed: determining an image frame of peak contrast by selecting the image frame with the greatest feature value, wherein the determined image frame of peak contrast is used to indicate the presence of the contrast agent in the main set of x-ray images, and wherein the semi-global optimization strategy comprises: performing a global search on a 2-D space of translation with a course grid using three different groups of five scale and rotation parameters to find a course-grid optimization; starting from the course-grid optimization, performing a global search with all five scale and rotation parameters within a region defined by a size of one grid of the course grid to enhance the course-grid optimization; and performing fine-tuning of the enhanced course-grid optimization using a hill climbing algorithm as a local optimization strategy within a region that is smaller than the size of one grid of the course grid, wherein the five scale and rotation parameters include x-direction scale, y-direction scale, x-direction translation, y-direction translation, and rotation.

12. The method of claim 11 , wherein the estimation of the background is updated as the main set of x-ray images are acquired based on one or more most recent image frames that are classified as not including contrast.

13. The method of claim 11 , wherein determining the measure of image intensity for each subtracted image includes generating a non-linear histogram mapping of each subtracted image.

14. The method of claim 13 , wherein selecting one or more of the subtracted images having a highest image intensity includes determining one or more highest histogram values.

15. The method of claim 11 , wherein the predefined shape model is modified in accordance with an anatomical structure of the subject prior to fitting.

16. The method of claim 11 , wherein the predefined shape model represents an aortic root.

17. The method of claim 11 , additionally comprising: selecting a set of contrast frames from among the plurality of image frames that have a feature value that is sufficiently close to the frame of peak contrast; selecting a set of non-contrast frames from among the plurality of image frames that have a feature value that is sufficiently far to the frame of peak contrast; training a local classifier using the selected set of contrast frames as positive training data and using the selected set of non-contrast frames as negative training data; and determining whether each of the frames of the plurality of image frames that are neither sufficiently close to the frame of peak contrast nor sufficiently far to the frame of peak contrast are contrast frames or non-contrast frames using the trained local classifier.

18. The method of claim 11 , additionally comprising registering a 3D image of an aortic root to the image frame determined to be of peak contrast and displaying the registered image.

19. The method of claim 18 , wherein the displayed registered image is used as visual guidance in performing an interventional procedure.